Assessing the nest building activity of ground-nesting bees by surveilling tumuli changes using camouflaged instance segmentation

Long-term monitoring of wild bees is essential to understand their biology and ecology. In ground-nesting bees, nesting activity is traditionally assessed by repeated counts of active nests, identifiable by the resulting piled-up soil above each nest entrance, known as tumulus. Counting nests manually is a time-consuming and complex task, which restricts the frequency and precision of observations. Additionally, the quantification, colourization and structure of excavated soil has so far been neglected as an informative metric of the bees’ nest building activity. Here, we present an automated approach to quantify nesting activity by counting tumuli and estimating the size of each tumulus using computer vision. Our method utilizes Camouflaged Instance Segmentation and can accurately distinguish individual tumuli from the background even when nests are closely spaced. We were able to assign tumuli segments to individual nest entrances, which allowed us to detect changes in the nest building activity over time. We present two exemplary descriptive analyses of the nest building activity both at population and at individual nest levels, demonstrating how changes in the total tumulus number and in the individual tumulus size could be linked to local weather conditions. Our method allows a standardized, non-invasive long-term observation of bee nesting activity with a high temporal resolution and low observer effort. In addition, we provide a dataset to support future evaluations of data-driven methods for segmentation and classification of ground-nesting bee nests, based on our model species Andrena vaga .